The Effect of the 2008 Great Recession on Child Poverty in California:

A Fixed Effects Model

Economics Honors Thesis Spring 2015

Laura Tsui

23601353

Advisor:

Professor Hilary Hoynes1

Abstract

This paper investigates how the 2008 financial crisis affected child poverty in California. Using panel data from 2003 to 2012, this paper evaluates the effect of unemployment and foreclosure rates, as

well as exogenous household characteristics, on the likelihood of a child being in poverty in California. Data from both the individual household level and the Metropolitan Statistical Area

(MSA) level are used. The main finding is that the recession has had a statistically significant effect on the likelihood of a child being in poverty in terms of unemployment and foreclosure rates, especially when fixed effects are applied. At the same time, household characteristics such as

education, race and marital status are also statistically significant in their effects on a child’s poverty status. However, changes in exogenous household characteristics can have a larger effect on child’s

poverty status as compared to the cyclicality of the economy.

                                                                                                               1  I would like to sincerely thank Professor Hilary Hoynes for her invaluable insights and advice the development of this paper. I would also like to thank Harrison Dekker from the UC Berkeley data lab for guiding me with his technical expertise.

   

1. Introduction: The impacts of the 2008 Global Recession remain evident even five years on, and children

throughout the United States continue to be negatively impacted. Children have the highest poverty

rates in the U.S. (Bitler, Hoynes and Kuka, 2014) and according to the Current Population Reports

from the Census, 24.3% of children in California are in poverty as of 2011. This statistic would

include another 1.3 million if it were not for safety net resources.

Poverty in the United States is measured with an “absolute” as opposed to a “relative”

measurement that is normally championed by other countries in the OECD. U.S. official poverty is

determined using total pre-tax family cash income relative to poverty thresholds that vary by family

size, the number of children and elderly persons. In 2012, the poverty threshold for a family of four

(two parents and two children) was $23.283 (Census Bureau, 2012). Child poverty rate is expressed

as the number of children who are in poverty as a percentage of the total number of children.

This research aims to evaluate the impact of the recession on child poverty in California

through unemployment rates and foreclosure rates, taking into account existing exogenous

demographic factors. This paper analyzes the effects on the likelihood of a child being in poverty

given a mix of endogenous variables from the financial crisis, as well as exogenous demographic

variables of an individual child’s household. The household-level factors in this research are: the

head of household’s age, marriage status, race and level of educational attainment. The macro factors

are the unemployment and foreclosure rates, the latter of which is also used as a proxy to estimate the

effect of the housing crisis. Foreclosure and unemployment rates are calculated on a Metropolitan

Statistical Area (MSA) level, a geographical region consisting of one or more counties that has a

relatively high population density at its core and close economic ties throughout the area. Since this

paper makes use of different sources that make use of different MSA groupings, I have regrouped the

MSAs to create a new combined list of 14 MSAs. This paper focuses on child poverty status on an

individual level, as opposed to the often-used child poverty rates.

   

However, the conclusion drawn from the poverty measure on a micro scale can also be

applied to child poverty on a macro measure, that is, child poverty rates at MSA level, since that is

the mean of the individual poverty status across the population. The regressions are weighted by

MSA, hence the OLS coefficient will be unbiased when applied to the mean.

The regressions were run on cross-sectional data in California across ten years: the first

subset being five years before the crisis (2003-2007), while the second subset would be the five years

inclusive of and after the crisis (2008-2012). The data was obtained from the Integrated Public Use

Microdata Series – Current Population Survey (IPUMS-CPS), RAND Organization California, as

well as the Bureau of Labor Statistics – Local Area Unemployment (BLS-LAU).

The data on a child’s poverty status was taken from IPUMS, which is a database from the

Census. This is the poverty status of a given child at an individual level. I also calculated child

poverty rates from IPUMS, and verified them with the MSA-level weighted data calculated from

county-level data taken from the Small Area Income and Poverty Estimates (SAIPE), which is also

from the Census. I took unemployment rates directly from the BLS and weighted them by MSAs.

Foreclosure rates were calculated using IPUMs and RAND data and were verified with the

information on the RealtyTrac and California Office of the Attorney General sites.

This paper reveals several interesting findings. First, as expected, the effects of

unemployment and foreclosure rates have been statistically significant in influencing a child’s

poverty status. Unemployment rates were used to measure the effect of the 2008 recession on child

poverty status, while the foreclosure rates were used as proxies for the housing crisis. When

unemployment rates rise, the average household income is likely to fall and the rate of child poverty

will rise. Foreclosures can also have a significant effect on child poverty, especially in California,

since the household may lose a significant portion of their assets, the children may be displaced and

the economy could be affected if construction had been a significant economic sector.

   

Secondly, the exogenous demographic factors are also statistically significant in influencing a

child’s poverty status on an individual level. Relative to a child whose head of household is white, a

child whose head of household is black is more likely to be in poverty, but less so if that person is

Asian. I also found that the higher the level of education of the head of household, the older he/she is,

and if he/she is married, then the child will be less likely to be in poverty.

When the recession variables were regressed together with the household characteristic

variables, the coefficients on both unemployment and foreclosure rates were statistically significant if

year-fixed effects were added. When adding MSA-fixed effects, unemployment rates remain

statistically significant but foreclosure rates are not. This suggests that the effect of unemployment

rates on child poverty are more greatly influenced by time when compared to foreclosure rates. When

testing both year and MSA-fixed effects jointly, the F-statistic was statistically significant, meaning

that they both have a joint effect on child poverty status.

2. Literature Review

While existing literature acknowledges the severity of the negative impact of the 2008

financial crisis in the US, there is only a modest amount of literature that directly addresses how the

crisis affects child poverty relative to the existing demographic characteristics of a household.

Instead, researchers have primarily focused on child poverty status in the United States as a whole,

overlooking many micro factors, such as the characteristics of the household, as well as geographical

inequalities.

A significant quantity of research discusses the severity of child poverty in California, but

only a modest number directly addresses the 2008 crisis as a direct cause. Using Census Population

data, Fuentes, O’Leary and Barba (2013) reports that currently, 23% of all children in California live

at or below the poverty line, and gathers that child poverty increased nearly by nearly 21% from

2006 – 2011. There is also great inequality between different races, counties and household types

(e.g. single motherhood versus dual-income families). Similarly, Isaacs (2011) notes that children in

   

the U.S. with unemployed parents suffer more severely than others. Bardhan and Walker (2010) and

Bohn (2011) find that single-female households suffer the most, with poverty rates at 35.4%

compared to 10.6% for dual income families. While these papers provide a strong foundation to the

context of the status of child poverty in the Golden State, they do not address the Great Recession as

a cause, which is what this paper seeks to achieve.

Other research papers evaluate the explanatory factors of child poverty in economic crises,

but either in a global or national context. Bitler, Hoynes and Kuka (2014) conduct analyses of the

impact of the recession on child poverty in the United states and find that child poverty rises in

recessions and falls in expansions, as most would expect. They construct two of their own measures

of child poverty, the after-tax and transfer income (ATTI), which adds to cash income the cash value

of non-cash programs like food stamps, less taxes and private income (PI) poverty which excludes all

government tax and transfer benefits. Both measures are constructed at household level. By running a

regression on state panel model and data from 1980 to 2012 they estimate the relationship between

the business cycle and child poverty. They find that the cyclicality of ATTI child poverty is

significantly weakened relative to the cyclicality of PI poverty, which demonstrates how the social

safety net insures vulnerable groups, and this has many public policy implications.

In exploring the effect of different household characteristics on child poverty, Litcher and

Eggebeen (1994) highlight the central role of parental employment in ameliorating poverty among

children, with poverty rates of children in working female-headed families being three times higher

than a female head that is unemployed. Race is also a crucial factor. They have also found, in another

1991 paper, that child poverty and racial inequality remain constant even in the context of changing

family structures in America. Furthermore, when comparing black and white families of equal

income or poverty status, white families tend to have higher assets (Oliver and Shapiro 2006) and

will generally live in areas that are usually middle-class.

   

There has also been literature that attempts to explain the inequities in child poverty through

the housing crisis. While children are often overlooked in housing industry statistics, a 2012 report

by Isaacs found that 1 in 10 children in the United States was affected by the foreclosure crisis in

2007, and these children had faced greater hardships and growing child poverty. Isaacs also estimates

that between 12 to 19 percent of foreclosures in California, between 2004 and 2008, had displaced

children, and can directly affect child poverty. Furthermore, Freelon, Betrand and Rogers (2012)

have cited particular hard-hit industrial sectors, including construction and housing, which were key

areas of California’s economic growth and are now particularly vulnerable to the economic

meltdown after the mortgage crisis, subsequently adversely affecting child poverty.

When comparing the effects of exogenous, micro-level factors and macro-level economic

conditions, there has been literature citing that the impact on child poverty from household

characteristics can be more significant. In a study about the effect of low income on child health,

Burgess, Propper and Rigg (2004) found that the direct impact of income is small. Rather, the

behavior of the household and decision-making, particularly the mother’s, is more significant. Blau

(1999) found that small impacts of short-term changes in income, as opposed to broader changes

from policy and economic conditions have some measurable effect on child development, but

household characteristics remain the most significant. Dahl and Lochner (2012) also found that,

when studying the relationship between household income and children’s test scores, while there can

be short term improvements in a child’s test scores with an increase in household income, there were

negligible long-term effects.

Most of the literature surveyed broadly fall into two categories: either they provide a

quantitative background to the child poverty in California, but do not offer a thorough evaluation of

the explanatory variables, or they explain causal variables but in a generic national context not in

relation to the 2008 Great Recession. This paper seeks to fill in that very gap by evaluating the

   

impact of the financial crisis on child poverty based on a comprehensive set of indicators, from

household-level demographic factors to MSA-level unemployment and foreclosure rates.

This paper will thus build on previous literature to create an analysis of child poverty trends

before and after the Great Recession that focuses on California. Recognizing the causes,

consequences, and welfare programs for child poverty, this paper will not attempt to “solve” the

problem, but may provide greater insights into effective policy planning for public agencies in

California for those who want to find a possible panacea to the growing problem of child poverty.

3. Data and Sample Description

Integrated Public Use Microdata Series – Current Population Survey (IPUMS-CPS)

I have obtained statistics for child poverty status as well as the exogenous characteristics of

the heads of household (race, marriage status, education, age) from the Integrated Public Use

Microdata Series – Current Population Survey (IPUMS-CPS), a data source of census microdata for

economic and social research. This is an integrated set of annual data from 50 years (1962-2011) of

the March Current Population Survey (CPS). The CPS is a monthly U.S. household survey conducted

jointly by the U.S. Census Bureau and the Bureau of Labor Statistics. In this IPUMS data set, the

unit of observation is the member of a household per year, over ten years.

However, the one standard sampling caveat in the use of IPUMS-CPS is that the IPUMS-

CPS samples are weighted since it involves a relatively smaller sample size as compared to sources

of aggregate data. Thus, sample weights need to be applied in order to obtain representative statistics

from the samples. In the case of this paper, I am conducting household-level analyses of March data,

and I will weight the households using the household-weight supplement variable that automatically

adjusts the sample.

Using the IPUMS data, I have also calculated child poverty rates per MSA, which are

weighted averages of the child poverty status variable on an MSA level.

RAND Organization

   

I obtained foreclosure data from RAND organization, which provides publicly available

economic data. The foreclosure numbers are sourced from DataQuick News, which ultimately

obtains its data from the Census. It has the total number of foreclosures by county, year and

Metropolitan Statistical Areas (MSA).

Bureau of Labor Statistics – Local Area Unemployment (BLS LAU)

I obtained the official unemployment rates by county from the BLS, which is taken from the

Census. However, since several counties are in non-metropolitan areas, and are hence not part of any

MSAs, I dropped them from the regression. For the remaining unemployment rates in the counties, I

merged them to fit the combined MSAs used in this paper. By weighting my regressions by the

population size of an MSA, I am able to generate coefficients for unemployment rates by MSA,

which would theoretically be given by the formula:

Description of Sample Data:

From the IPUMS Dataset, there are 185, 255 observations, one per person per year, over a

period of ten years from 2003 – 2012. Among that number, 55,647 are children who are ages 0 to 17,

since that is the standard definition of a child in all statistics. I dropped those who are not in any

metropolitan areas, and are therefore not part of any MSAs, since they will not be covered in my

regression. This is because foreclosure rates from RAND are available only at the MSA level, and

the sample size will also be larger for each MSA as opposed to counties.

I obtained the total number of foreclosures in a year from RAND from all available counties

that are metropolitan. I construct foreclosure rates in an MSA to be the count of the total number of

foreclosures divided by the weighted number of households in the MSA, which in turn is taken from

IPUMS. Since the number of households change every year, calculating foreclosure rates with a

different denominator every year would not be able to give us an accurate understanding of the

   

relationship between foreclosure rates and child poverty. Hence I used 2003 as the base year to

calculate the foreclosure rates as follows:

Similarly, I used the given unemployment rates taken from the BLS LAU and merged them

with the other two data sets. Another key change I made to the data sets is the recombination of

different counties to form new combined MSAs. Please see table 1 in the Description of Variables

section for more details.

Description of Dependent Variable:

In this paper, I have only one dependent variable, which is child poverty status.

-­‐ CHILDPOV – Child Poverty Status Dummy (1,0). Using the official IPUMS poverty

variable, which used by the CPS is the same as that of the U.S. Census Bureau and is based

on the 1961 economy food plan. For the relevant years of this project (2003 – 2012), the data

on IPUMS-CPS covers all individuals of all ages. I have defined children as those who are

under 18 years old, which is generally the convention when defining children.

Description of Independent variables:

-­‐ HH EDUC – Educational Attainment Dummy (0,1) of the Head of Household. There are

four dummies, each for different levels of educational attainment

o HHEDUC1 – Less than 12 years of schooling

o HHEDUC2 – 12 years of schooling/High School Diploma

o HHEDUC3 – Some college

o HHEDUC4 – College or more (Tertiary education)

-­‐ HH RACE – Race Dummy (0,1) of the head of household. I grouped the races into 4

different categories for ease of understanding the data.

   

o HHRACE1 – Whites

o HHRACE2 – Blacks

o HHRACE3 – Asian/Pacific Islander

o HHRACE 4 – Other

-­‐ HMARRIED: Marriage dummy. I created a binary variable for those whose head of

household are married

-­‐ HAGE: Age of the head of household.

MSA: Compared to counties, MSAs have a wider regional coverage and can allow for more

accurate analysis. Notably, the RAND data set had different MSAs from the IPUMs data set,

and I regrouped the counties based on geographical proximity in order to create the new data

set, giving 14 new MSAs. I have also dropped those that are in IPUM’s “Not In MSA” group.

In my regression, the numbers of the MSAs correspond in the order in which they are

presented in Table 2.

   

.

Total 186,255 100.00 Riverside-San Bernardino 19,635 10.54 100.00 Ventura-Oxnard-ThousandOaks 4,292 2.30 89.46 Stockton 3,164 1.70 87.15 Santa Rosa-Petaluma 2,040 1.10 85.45 Santa Barbara-Santa Maria-Lompoc 1,963 1.05 84.36 San Jose-Santa Cruz 10,869 5.84 83.31 San Francisco-Oakland-Vallejo 22,939 12.32 77.47San Diego-Carlsband-San Marcos-ElCentro 15,585 8.37 65.15 Salinas-Monterey-SanLuisObispo 3,454 1.85 56.79 Sacramento-Yolo-Yuba City 12,887 6.92 54.93 Modesto 2,799 1.50 48.01 LosAngeles-LongBeach 75,128 40.34 46.51 Fresno-Madera-Merced 7,380 3.96 6.17 Bakersfield 4,120 2.21 2.21 RECODE of metarea (Metropolitan area) Freq. Percent Cum.

Table  1  Combining  different  MSAs  from  IPUMS  and  RAND  

Table  2  Combined  MSAs  and  corresponding  numbers    

   

-­‐ Foreclosure Rates: Calculated as the total number of foreclosures divided by the total

weighted number of households in each MSA.

-­‐ Unemployment Rate: Calculated as the total number of unemployed individuals divided by

the labor force.

4. Model Description

There are a total of 11 regressions performed in this paper. The first four will focus on the effect of

endogenous recession variables on child poverty, while the next will estimate the effects of

exogenous household characteristics, and finally all the variables will be regressed together. All of

the models look only at the child population, and hence will have only 55,647 observations.

Models (1) to (4): Regression of child poverty status on recession variables

CHILDPOV i,t = β0 + β1 URm,t + ɛi,t -----(1)

CHILDPOV i,t = β0 + β1 FCRm,t + ɛi,t -----(2)

CHILDPOV i,t = β0 + β1 URm,t + β2FCR m,t + ɛi,t -----(3)

CHILDPOV i,t = β0 + β1 URm,t + β2FCR m,t + β11MSA2 + … + β24MSA14 + β25YEAR2004 + … +

β34YEAR2012 + ɛi,t -----(4)

Since this research aims to evaluate the effect of the financial crisis on child poverty, I have

regressed the endogenous recession variables, one at a time, and then together and with fixed effects.

Each of the first four models is intended to give us an idea of the effect that each recession variable,

i.e. unemployment and foreclosure rates, has on child poverty, and how the variations in geography

and time affect the results.

Models (5) to (11): Regression of child poverty status on household and recession variables

   

Model 5: Regression on head of household’s exogenous characteristics

CHILDPOVi,t = β0 + β1 HHRACE2i,t + β2 HHRACE3i,t + β3HHRACE4i,t + β4 HHMARRIEDi,t+ β5

HHEDUC1i,t + β6 HHEDUC2i,t+ β7HHEDUC3i,t + β8 HHAGEi,t + ɛi,t -----(5)

This model estimates the probability of a child being in poverty given his/her head of

household’s exogenous characteristics, which are race, highest level of education and age. I dropped

the white dummy as well as the ‘12 years of school’ education dummy variables to prevent

multicollinearity.

Ex Ante, I would expect that if a head of household is black or other, the child would be

more likely to be in poverty, while the reverse is true if the head of household is Asian. Since I

dropped the ‘college’ dummy, this means that all the other variables should have positive coefficients,

since fewer years of education for the head of household would mean that the child is more likely to

be in poverty.

Model (6): Estimating the effects of unemployment rates on child poverty

CHILDPOVi,t = β0 + β1 HHRACE2 i,t + β2 HHRACE3 i,t + β3HHRACE4 i,t + β4 HHMARRIED i,t + β5

HHEDUC1 i,t + β6 HHEDUC2 i,t + β7HHEDUC3 i,t + β8 HHAGE i,t + β9 UR m,t + ɛi,t -----(6)

This model builds on model 1, and therefore also includes the original exogenous

characteristics of the head of household. Using a panel data set of unemployment rates across

different MSAs over the decade, I regressed child poverty rates in the MSAs over the years on the

unemployment rates. To prevent multicollinearity, I dropped the White, high school degree, MSA1

(Bakersfield) and Year 2003 dummies. I also clustered the standard errors by MSA.

I would expect the unemployment rate to be positively correlated with the child poverty rate,

and for that coefficient to be statistically significant. This allows us to estimate and compare between

   

the effects of unemployment rates relative to existing household characteristics.

Model (7): Estimating the effects of foreclosure rates on child poverty (no unemployment)

CHILDPOV i,t = β0 + β1 HHRACE2 i,t + β2 HHRACE3 i,t + β3HHRACE4 i,t + β4 HHMARRIED i,t + β5

HHEDUC1 i,t + β6 HHEDUC2 i,t + β7HHEDUC3 i,t + β8 HHAGE i,t

+ β9 FCR m,t + ɛi,t -----(7)

This is a modification of the previous model. I ran the regression on foreclosure rates, and

dropped the same dummies to prevent multicollinearity. This model allows us to isolate the effect of

foreclosure rates in different MSAs on child poverty rates relative to household factors.

Model (8): Regression on both household-level and MSA-level factors

CHILDPOV i,t = β0 + β1 RHHRACE2 i,t + β2 HHRACE3 i,t + β3HHRACE4 i,t + β4 HHMARRIED i,t + β5

HHEDUC1 i,t + β6 HHEDUC2 i,t + β7HHEDUC3 i,t + β8 HHAGE i,t + β9 UR m,t

+ β10FCR m,t + ɛi,t -----(8)

This model combines the previous two models, allowing us to see the combined effect of

both unemployment and foreclosure rates, as well as household characteristics, on child poverty rates.

From the literature, we might expect the effect of recession variables to be either less statistically

significant, or to be of a smaller magnitude, hence having less of an effect on child poverty relative to

household characteristics.

Model (9) to (11): Regression of all independent variables and fixed effects

CHILDPOV i,t = β0 + β1 RHHRACE2 i,t + β2 RACE3 i,t + β3RACE4 i,t + β4 MAR i,t + β5 HEADEDUC1

i,t + β6 HEADEDUC2 i,t + β7HEADEDUC3 i,t + β8 HEADAGE i,t + β9 UR m,t + β10FCR m,t + β11MSA2

   

+ … + β24MSA14 + β25YEAR2004 + … + β34YEAR2012 + ɛi,t -----(11)

These models add year and MSA fixed effects one at a time, then both together, to model (8),

allowing us to see how periodic and geographic variations can have effects on child poverty status in

California.

5. Results:

Trends:

   

 

Figure  1  Child  Poverty  Rates  in  California.  Source:  Author’s  calculations  from  IPUMS-­‐CPS                                

       

Figure  2  Child  Poverty  Rates  in  key  MSAs.  Source:  Author’s  calculations  from  IPUMS-­‐CPS  

   

   

Figure  3  Child  Poverty  Rates  2003-­‐2007  Source:  Author’s  calculations  from  IPUMS-­‐CPS                                        

   

Figure  4  Child  Poverty  Rates  2008-­‐2012  Source:  Author’s  calculations  from  IPUMS-­‐CPS  

   

 Figure  5  Unemployment  Rates  in  California.  Source:  BLS  LAU                      

   

Figure  6  Unemployment  Rates  in  key  MSAs.  Source:  Author’s  calculations  from  BLS  LAU  

   

Figure  7  Unemployment  Rates  2003-­‐2007.  Source:  Author’s  calculations  from  BLS  LAU              

   

Figure  8  Unemployment  Rates  2008-­‐2012  Source:  Author’s  calculations  from  BLS  LAU  

     

 Figure  9  Foreclosure  Rates  in  California.  Source:  Author’s  calculations  from  RAND  and  IPUMS              

 

Figure  10  Foreclosure  Rates  in  California.  Source:  Author’s  calculations  from  RAND  and  IPUMS  

   

   

Figure  11  Foreclosure  Rates  2003-­‐2007.  Source:  Author’s  calculations  from  RAND  and  IPUMS            

 

Figure  12  Foreclosure  Rates  2008-­‐2012.  Source:  Author’s  calculations  from  RAND  and  IPUMS  

   

Child poverty rates in California have been on the rise since its low point in 2007, at a rate

of over 18%. Since then, child poverty has increased to a height of over 24% in 2012, with the

sharpest rate of increase between 2010 and 2011. However, when we look at the child poverty rates

in different MSAs, the trend is less clear. In fact, many of the larger MSAs, such as Los Angeles-

Long Beach or Riverside-San Bernardino have seen relatively stable trends, or a slight increase

towards 2012. On the other hand, poorer MSAs in the Central Valley, such as Modesto, have seen

sharp increases in child poverty, with a sharp rise leading up to the crisis in 2007-2008. The increases

in child poverty rates are highly unequal within the state of California, suggesting that the average

rate in California was pulled up because of poorer regions like Modesto. When mapping out the

average poverty rates, we can see that poverty rates are indeed highly uneven. On the legends for the

maps in Figures 3 and 4, we can see that overall, poverty rates have risen, especially in key areas.

Unlike the trends in child poverty rates, weighted unemployment rates in different MSAs

share the same trend as that of California as a whole. Unemployment had dipped to a low of about

6% in 2006, but rose to a peak in 2010 to over 13%, before dropping slightly to under 12% by the

end of 2012. There are similar but less pronounced trends in the different MSAs, but the trend is less

erratic as compared to that of child poverty rates. We also see, from the choropleth maps, that there is

a high level of unemployment in central California, but unemployment has risen across the board.

Foreclosure rates rose to a peak of about 0.19% in 2008, before settling to about 0.09% in

2012. Like unemployment rate trends, there are similar trends between foreclosure rates at a state

level and at the MSA level. Southern California areas, including Riverside-San Bernardino and Los

Angeles-Long Beach have the highest rates of foreclosure rates since the recession. Notably, these

areas with highest levels of foreclosure rates have rates that are much higher than that of California

as a whole (4.7% compared to 0.09%). We can also see the variation in the choropleth map.

   

Regression results:

(1) Child Poverty Status

(2) Child Poverty Status

(3) Child Poverty Status

(4) Child Poverty Status

urURURURUUdsUR 0.003** 0.003 -0.008* (%) (0.001) (0.002)

(0.004)

FCR 0.015*** 0.002 0.016*** Standard Errors Year FEs MSA FEs Mean of Dependent Var

Clustered No No 0.206

(0.004) Clustered No No 0.206

(0.05) Clustered No No 0.206

(0.03) Clustered Yes Yes 0.206

Constant 0.195*** (0.011)

0.197*** (0.008)

0.196*** (0.011)

0.320*** (0.048)

R2 0.00

0.00 0.00 0.01

N 55,647 55,647 55,647 55,647 Table 3. Results of the four regressions of Child Poverty Status on unemployment and foreclosure rates with fixed effects. Standard errors in parentheses. Clusters at the MSA level and by year. Mean of Dependent Variable is weighted by MSA. *p <0.10, * * p <0.05, * * * p < 0.01 Models (1) and (2) follow in line with my original expectations. When regressing child

poverty status against unemployment and foreclosure rates, we see that they are both statistically

significant and have a positive correlation with child poverty. A rise in unemployment rate

corresponds to a rise in 0.3 percentage points in the likelihood of a child being in poverty. Similarly,

from (2), we see that an increase in foreclosure rate leads to an increase of 1.5 percentage points in

the likelihood of a child being in poverty. Interestingly, when you combine the two variables together,

they are both statistically insignificant unless fixed effects are added, in which case both are

statistically significant. This could mean that the geographical and periodical variation affects how

the MSA-level unemployment and foreclosure rates affect child poverty status in these regions.

   

(5) Child Poverty

Status

(6) Child Poverty

Status

(7) Child Poverty

Status

(8) Child Poverty

Status

(9) Child Poverty

Status

(10) Child Poverty

Status

(11) Child Poverty

Status

Black 0.027*** 0.009 0.009 0.009 0.009 0.012 0.013 (0.008) (0.017)

(0.017) (0.017) (0.017) (0.017) (0.018)

Asian -0.074*** -0.089*** -0.089*** -0.089*** -0.089*** -0.080*** -0.080*** (0.005) (0.009)

(0.009) (0.009) (0.009) (0.010) (0.010)

Other -0.007 0.006 0.007 0.006 0.004* 0.011 0.009

(0.010) (0.022) (0.021) (0.022) (0.022) (0.022) (0.022)

<12 years 0.048*** 0.048 0.048*** 0.048*** 0.048*** 0.046*** 0.046***

(0.006) (0.015) (0.015) (0.015) (0.014) (0.015) (0.014)

Some -0.009** -0.013*** -0.013 -0.013 -0.012 -0.011 -0.010 Coll.

(0.005) (0.010) (0.010) (0.010) (0.010) (0.010) (0.010)

More than -0.022*** -0.020 -0.020 -0.020 -0.019 -0.017 -0.017 Coll.

(0.004) (0.012) (0.012) (0.012) (0.012) (0.012) (0.012)

Age -0.000*** -0.001 -0.001** -0.001** -0.000* -0.001*** -0.000*

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)

Married -0.220*** -0.206*** -0.206*** -0.206*** -0.205*** -0.205*** -0.204***

(0.004) (0.013) (0.013) (0.013) (0.013) (0.013) (0.013)

UR 0.003** 0.003 -0.004*** 0.006*** -0.010** (%)

(0.001) (0.001) (0.001) (0.002) (0.004)

FCR 0.006 -0.000 0.015*** -0.011 0.014*** (%)

Standard

Errors

Year FEs

MSA FEs

Mean of Dependent Variable

Not Clustered

No

No

0.206

Clustered

No

No

0.206

(0.003)

Clustered

No

No

0.206

(0.005)

Clustered

No

No

0.206

(0.004)

Clustered

Yes

No 0.206

(0.008)

Clustered

No

Yes

0.206

(0.004)

Clustered

Yes

Yes

0.206

Constant 0.399*** 0.383*** 0.402*** 0.383*** 0.435*** 0.367*** 0.503*** (0.007)

(0.018) (0.017) (0.018) (0.019) (0.031) (0.050)

R2 0.07

0.06 0.06 0.06 0.06 0.06 0.07

N 55,647 55,647 55,647 55,647 55,647 55,647 55,647 Table 4. Results of the seven regressions of child poverty status on household and recession variables. All household independent variables are characteristics of the head of household of the child. Total number of observations is the total number of children across the 14 MSAs and 10 years. Standard errors in parentheses. Clusters at the MSA level and by year. Dependent variable is weighted by MSA. *p <0.10, * * p <0.05, * * * p < 0.01

   

The results of Model (5) are also as expected. In comparison to children whose heads of households

are whites (since the white dummy was dropped), those whose heads of households are black are 3

percentage points more likely to be in poverty, but 7 percentage points less likely to be in poverty. In

all seven regressions, the ‘other’ race variable is statistically insignificant while the ‘Asian’ variable

is always statistically significant.

In terms of the educational attainment of the head of household, we see that, as

predicted, a child is less likely to be in poverty if the head of household has had more years of

education. Compared to a head of household with a high school degree, someone with less than 12

years of education would have a child who is about 5 percentage points more likely to be poor, but

someone with some college education would have a child who is about 1 percentage point less likely

to be poor. Notably, however, only the ‘<12 years’ education variable is statistically significant

across all seven regressions. This means that the effect of not completing a high school degree

compared to having a high school degree is has a more significant effect relative to having pursued

higher education compared to having a high school degree.

The older the head of household is in general, the less likely the child is to be in

poverty. This is expected, since the older the head of household, the more likely the person is

employed or has an income, and the less likely the household, including the child, will be poor.

In terms of marriage, if a head of household is married, the child will be about 20

percentage points less likely to be in poverty. This could be because the head of household married

another income earner and could pool their financial resources. Another reason could be that a person

who wants to marry is already financially stable.

When we factor in unemployment rates and foreclosure rates on a MSA level in

models (6) through (7), we see that some of the demographic factors of the head of household remain

statistically significant in influencing a child’s poverty status at an individual level, such as the Asian

race, a ‘<12 years’ of education, age and marriage status. This makes sense, since we are estimating

   

child poverty at an individual level, which is highly influenced by the head of household’s existing

characteristics, which can affect a child’s resistance to poverty and the cyclicality of the economy.

When we regress the recession variables together with the exogenous household

characteristics, we see that unemployment rates remain statistically significant except when year

fixed effects are not added. Otherwise, it remains highly significant at the 1% level and the

magnitude of the coefficient is not too different from models (1) – (4) when household variables

were not regressed. The effect of unemployment rates on child poverty is hence also influenced by

the year. This suggests that the crisis did indeed have a statistically significant impact on the

likelihood of a child being in poverty, since the year fixed effects are also proxies for cyclicality.

Foreclosure rates are also statistically significant once year fixed effects are regressed,

possibly because the foreclosure crisis led to a sharp increase in foreclosure rates relative to the pre-

crisis period. Similar to the coefficient on unemployment rates, the coefficient of foreclosure rates

does not alter very much when head of household characteristics are added. The fact that the addition

of year fixed effects can affect the statistical significance of both recession variables suggest that

there is a strong macro trend for California in terms of child poverty.

In order to see the significance of both the two fixed effects, I jointly tested them and

found that the F-statistic was statistically significant. Using model (11) which regresses all the

different macro and micro variables, I first jointly tested the year dummies to find that there was a

statistically significant effect of time on child poverty status (F(9, 139) = 13.97, P = 0.000). Next,

I jointly tested the MSA dummies to find a statistically significant effect of geographical variation on

child poverty status (F(13, 139) = 28.32, P = 0.000). Finally, I jointly tested all time and

geographical fixed effects, finding a statistically significant effect of both on child poverty status

(F(22, 139) = 27.12, P = 0.000).

From the regressions, we find that both recession and exogenous variables have

statistically significant influences on a child’s poverty status. Notably, however, some head of

   

household characteristics are always statistically significant, and the magnitude of their coefficients

is also larger when compared to that of unemployment and foreclosure rates. For example, if the head

of household is married, the effect is a 20 percentage point deduction in the likelihood of a child

being in poverty, as opposed to a 0.3 percentage point difference from the impact of MSA level

unemployment rate. This means that changing certain demographic characteristics of a head of

household has a greater impact on a child’s poverty status than macro, financial crisis factors.

6. Conclusion

The analysis of factors associated with childhood poverty in California and the recession are

complex, involving social issues, educational and economic achievement. This paper aimed to show

the effect of the 2008 Great Recession on child poverty in California, and measures the effect of the

external crisis on children relative to the existing exogenous characteristics of the head of household.

In particular, I found that on an individual level, while both economic and household characteristics

are statistically significant in affecting a child’s poverty status, the effect of a change in demographic

factors can have a much greater impact on the poverty status.

This can have several policy implications. Firstly, it means that improving the demographic

conditions of a household, such as a head of household’s education, can have significant effects on a

child’s poverty rate. This means that investing in a child’s education now can have significant

impacts on the family in future when, in turn, the child becomes an adult. Secondly, it also means

that inequality is also a significant factor in influencing California’s overall child poverty rates, hence

targeted policies in specific counties and MSAs would be most effective.

   

7. Bibliography:

1. Bardhan, A., and Walker, R. (2010) ‘California, pivot of the Great Recession’. Working

paper No. 203-10, Institute for Research on Labor and Employment, U.C. Berkeley.

2. Bitler, Marianne, Hoynes, Hilary and Kuka, Elira (November 2014), “Child Poverty and the

Great Recession in the United States”, Working Paper Under Review, UNICEF Report on

Child Poverty and the Great Recession

3. Blau, David M., (May 1999) ‘The Effect of Income on Child Development’, The Review of

Economics and Statistics of MIT, Vol. 81, No. 2, Pages 261-276

4. Burgess, Simon, Propper, Carol, and Rigg, John (2004), ‘The Impact of Low-Income on

Child Health: Evidence from a Birth Cohort Study’, The University of Bristol Leverhulme

Center for Market and Public Organization Working Paper Series, No. 04/098, retrieved

May 6, 2015, from: http://www.bristol.ac.uk/media-

library/sites/cmpo/migrated/documents/wp98.pdf

5. Bohn, Sara (2011) “Child Poverty in California,” Public Policy Institute of California,

[available at http://www.ppic.org/content/pubs/jtf/ JTF_ChildPoverty.pdf]

6. Dahl, Gordon B., and Lochnew, Lance (2012), ‘The Impact of Family Income on Child

Achievement: Evidence from the Earned Income Tax Credit.” American Economic Review,

102(5): 1927-56, 10.1257/aer.102.5.1927

7. Fuentes, Rey, O’Leary, Ann, Barba, James (2013), ‘Prosperity Threatened: Perspectives on

Childhood Poverty in California’, The Center for the Next Generation January Issue

8. Isaacs, Julia B., (December 2011), The Recession’s Ongoing Impact on America’s Children:

Indicators of Children's Economic Well-Being Through 2011, The Brookings Institution

retrieved May 6, 2015, from:

   

http://www.usatoday.com/story/money/business/2013/09/14/impact-on-states-of-2008-

financial-crisis/2812691/

9. Isaacs, Julia B., (March 2012 ‘The Ongoing Impact of Foreclosures on Children’, ) The

Brookings Institution, retrieved May 6, 2015, from:

http://www.brookings.edu/research/papers/2012/04/18-foreclosures-children-isaacs

10. Eggebeen, David and Litcher, Daniel (1994), ‘The Effect of Parental Employment on Child

Poverty’, Journal of Marriage and Family, Vol. 56, No. 3 (Aug., 1994), pp. 633-645

11. Eggebeen, David and Lichter, Daniel (1991), ‘Race, Family Structure, and Changing Poverty

among American Children’, American Sociological Review, Vol. 56, No. 6, pp. 801 – 815

12. Oliver, Melvin and Shapiro, Thomas (2006) ‘Black Wealth/ White Wealth: A New

Perspective on Racial Inequality’, Routledge Publishing